FlyTracker

Technical report: Coming soon!

Input: FlyTracker accepts any standard video format, as well as ufmf and seq format.

For best results, make sure that:

background is fixed
chamber and camera are rigid during filming
lighting is such that flies do not cast shadows and light does not reflect off chamber
spatial resolution is high enough

tested on 3-30 pixels per mm

for wing detection: >= 7 PPM

for leg detection: >= 12 PPM

temporal resolution is high enough

tested on 25-200 frames per second

some fly-actions take less than 30 ms (~1 frame at 30 FPS, 6 frames at 200 FPS)

We recommend that you iteratively apply the FlyTracker to test your setup while developing it.
For reference, check out our sample videos.

Output:

-track.mat	raw tracking data (e.g. position, orientation, left wing angle)
trk.names	{1 x n_fields cell} names of fields in trk.data
trk.data	[n_flies x n_frames x n_fields double] tracking data matrix
trk.flags	[n_flags x 5 double] potential identity swaps (fly1 fly2 start_fr end_fr ambig)
-feat.mat	features derived from tracking data (e.g. velocity, distance to wall)
feat.names	{1 x n_fields cell} names of fields in feat.data
feat.units	{1 x n_fields cell} units of fields in feat.data
feat.data	[n_flies x n_frames x n_fields double] feature data matrix
-seg.mat	foreground/body/wing/leg pixel locations
-JAABA	folder suitable for JAABA processing, containing tracking and video data
-actions.mat	list of actions annotated by user using the visualizer tool
behs	{1 x n_behs cell} names of annotated actions
bouts	{n_flies x n_behs cell} list of bouts for all flies and actions (start_fr end_fr certainty)

Feature description:

individual features
'vel'
'ang_vel'
'min_wing_ang'
'max_wing_ang'
'mean_wing_length'
'axis_ratio'
'fg_body_ratio'
'contrast'
'dist_to_wall'
relative features
'dist_to_other'
'leg_dist'
'angle_between'
'facing_angle'

speed of fly
angular velocity of fly
mininum of left and right wing angle
maximum of left and right wing angle
mean of left and right wing length
major axis length divided by minor axis length
foreground area divided by body area
mean gradient in image around fly
distance to wall

distance between the centers of the two flies
minimum distance between foregrounds of the two flies
angle between the facing vectors of the two flies
angle between facing vector and the vector towards other fly

Caltech FlyTracker